Akira Hatanaka [Wed, 17 Feb 2016 21:09:50 +0000 (21:09 +0000)]
[CodeGen] Fix an assert in CodeGenFunction::EmitFunctionEpilog
The assert is triggered because isObjCRetainableType() is called on the
canonicalized return type that has been stripped of the typedefs and
attributes attached to it. To fix this assert, this commit gets the
original return type from CurCodeDecl or BlockInfo and uses it instead
of the canoicalized type.
rdar://problem/
24470031
Differential Revision: http://reviews.llvm.org/D16914
llvm-svn: 261151
Alexey Samsonov [Wed, 17 Feb 2016 21:00:50 +0000 (21:00 +0000)]
Fix PR26608: Make sanitizer_common tests more portable.
llvm-svn: 261150
Haicheng Wu [Wed, 17 Feb 2016 21:00:06 +0000 (21:00 +0000)]
[LIR] Avoid turning non-temporal stores into memset
This is to fix PR26645.
llvm-svn: 261149
Alexey Samsonov [Wed, 17 Feb 2016 20:40:10 +0000 (20:40 +0000)]
[TSan] PR26609: Fix two test cases.
llvm-svn: 261148
Adrian Prantl [Wed, 17 Feb 2016 20:02:25 +0000 (20:02 +0000)]
Debug Info: Teach LdStHasDebugValue() (Local.cpp) about DIExpressions.
This function is used to check whether a dbg.value intrinsic has already
been inserted, but without comparing the DIExpression, it would erroneously
fire on split aggregates and only the first scalar would survive.
Found via http://reviews.llvm.org/D16867.
<rdar://problem/
24456528>
llvm-svn: 261145
George Burgess IV [Wed, 17 Feb 2016 19:59:32 +0000 (19:59 +0000)]
Add static/const qualifiers to methods. NFC.
Split out this change as requested in D14933.
llvm-svn: 261144
Kostya Serebryany [Wed, 17 Feb 2016 19:42:34 +0000 (19:42 +0000)]
[libFuzzer] don't timeout when loading the corpus. Be a bit more verbose when loading large corpus.
llvm-svn: 261143
Alexey Samsonov [Wed, 17 Feb 2016 19:35:51 +0000 (19:35 +0000)]
[tests] Slightly improve a fix in r260669.
llvm-svn: 261142
Akira Hatanaka [Wed, 17 Feb 2016 19:35:47 +0000 (19:35 +0000)]
Mention 'notail' attribute in 3.9 release notes.
llvm-svn: 261141
Elena Demikhovsky [Wed, 17 Feb 2016 19:23:04 +0000 (19:23 +0000)]
Create masked gather and scatter intrinsics in Loop Vectorizer.
Loop vectorizer now knows to vectorize GEP and create masked gather and scatter intrinsics for random memory access.
The feature is enabled on AVX-512 target.
Differential Revision: http://reviews.llvm.org/D15690
llvm-svn: 261140
Amaury Sechet [Wed, 17 Feb 2016 19:21:28 +0000 (19:21 +0000)]
Fix load alignement when unpacking aggregates structs
Summary: Store and loads unpacked by instcombine do not always have the right alignement. This explicitely compute the alignement and set it.
Reviewers: dblaikie, majnemer, reames, hfinkel, joker.eph
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D17326
llvm-svn: 261139
David Majnemer [Wed, 17 Feb 2016 19:02:36 +0000 (19:02 +0000)]
Revert "Reapply commit r258404 with fix."
This reverts commit r259357, it caused PR26629.
llvm-svn: 261137
Ed Schouten [Wed, 17 Feb 2016 18:56:20 +0000 (18:56 +0000)]
Enable SafeStack for CloudABI.
Summary:
I've got a patchset in my home directory to integrate support for
SafeStack into CloudABI's C library. All of the CloudABI unit tests
still seem to pass. Pretty sweet!
This change adds the necessary changes to Clang to make
-fsanitize=safe-stack work on CloudABI. Without it, passing this command
line flag throws an error.
Reviewers: eugenis, samsonov
Differential Revision: http://reviews.llvm.org/D17243
llvm-svn: 261135
Frederic Riss [Wed, 17 Feb 2016 18:51:27 +0000 (18:51 +0000)]
[ObjCARC] Handle ARCInstKind::ClaimRV in OptimizeIndividualCalls.
When support for objc_unsafeClaimAutoreleasedReturnValue has been added to the
ARC optimizer in r258970, one case was missed which would lead the optimizer
to execute an llvm_unreachable. In this case, just handle ClaimRV in the same
way we handle RetainRV.
llvm-svn: 261134
Colin LeMahieu [Wed, 17 Feb 2016 18:50:21 +0000 (18:50 +0000)]
[Hexagon] Replacing reference/dereference with reference cast.
llvm-svn: 261133
Nico Weber [Wed, 17 Feb 2016 18:48:08 +0000 (18:48 +0000)]
Remove superfluous semicolon.
llvm-svn: 261128
Nico Weber [Wed, 17 Feb 2016 18:47:29 +0000 (18:47 +0000)]
Revert r261070, it caused PR26652 / PR26653.
llvm-svn: 261127
David Majnemer [Wed, 17 Feb 2016 18:37:11 +0000 (18:37 +0000)]
[WinEH] Optimize WinEH state stores
32-bit x86 Windows targets use a linked-list of nodes allocated on the
stack, referenced to via thread-local storage. The personality routine
interprets one of the fields in the node as a 'state number' which
indicates where the personality routine should transfer control.
State transitions are possible only before call-sites which may throw
exceptions. Our previous scheme had us update the state number before
all call-sites which may throw.
Instead, we can try to minimize the number of times we need to store by
reasoning about the nearest store which dominates the current call-site.
If the last store agrees with the current call-site, then we know that
the state-update is redundant and can be elided.
This is largely straightforward: an RPO walk of the blocks allows us to
correctly forward propagate the information when the function is a DAG.
Currently, loops are not handled optimally and may trigger superfluous
state stores.
Differential Revision: http://reviews.llvm.org/D16763
llvm-svn: 261122
Ed Maste [Wed, 17 Feb 2016 18:25:27 +0000 (18:25 +0000)]
[tsan] Fix signal number definitions for FreeBSD
The change in r253983 for OS X also applies to FreeBSD.
llvm-svn: 261121
Ed Maste [Wed, 17 Feb 2016 18:22:50 +0000 (18:22 +0000)]
[tsan] Fix build warnings on FreeBSD
The change in r252165 for OS X applies to FreeBSD as well.
llvm-svn: 261120
Easwaran Raman [Wed, 17 Feb 2016 18:18:47 +0000 (18:18 +0000)]
Add a profile summary class specific to instrumentation profiles.
Modify ProfileSummary class to make it not instrumented profile specific.
Add a new InstrumentedProfileSummary class that inherits from ProfileSummary.
Differential Revision: http://reviews.llvm.org/D17310
llvm-svn: 261119
Colin LeMahieu [Wed, 17 Feb 2016 18:14:05 +0000 (18:14 +0000)]
[Hexagon] Loop instructions don't need special processing. Extension and fitting is performed by generic code and the comment is incorrect, loops don't have a separate extended opcode.
llvm-svn: 261118
Justin Lebar [Wed, 17 Feb 2016 17:46:54 +0000 (17:46 +0000)]
[NVPTX] Annotate convergent intrinsics as convergent.
Summary:
Previously the machine instructions for bar.sync &co. were not marked as
convergent. This resulted in some MI passes (such as TailDuplication,
fixed in an upcoming patch) doing unsafe things to these instructions.
Reviewers: jingyue
Subscribers: llvm-commits, tra, jholewinski, hfinkel
Differential Revision: http://reviews.llvm.org/D17318
llvm-svn: 261115
Justin Lebar [Wed, 17 Feb 2016 17:46:52 +0000 (17:46 +0000)]
[NVPTX] Test that MachineSink won't sink across llvm.cuda.syncthreads.
Summary:
The syncthreads MI is modeled as mayread/maywrite -- convergence doesn't
even come into play here. Nonetheless this property is highly implicit
in the tablegen files, so a test seems appropriate.
Reviewers: jingyue
Subscribers: llvm-commits, jholewinski
Differential Revision: http://reviews.llvm.org/D17319
llvm-svn: 261114
Justin Lebar [Wed, 17 Feb 2016 17:46:50 +0000 (17:46 +0000)]
[NVPTX] Annotate call machine instructions as calls.
Summary:
Otherwise we'll try to do unsafe optimizations on these MIs, such as
sinking loads below calls.
(I suspect that this is not the only bug in the NVPTX instruction
tablegen files; I need to comb through them.)
Reviewers: jholewinski, tra
Subscribers: jingyue, jhen, llvm-commits
Differential Revision: http://reviews.llvm.org/D17315
llvm-svn: 261113
Justin Lebar [Wed, 17 Feb 2016 17:46:47 +0000 (17:46 +0000)]
[IR] Add {is,set,setNot}Convergent() functions to CallSite, CallInstr, and InvokeInstr.
Summary:
(CallSite already has isConvergent() and setConvergent().)
No functional changes.
Reviewers: reames
Subscribers: llvm-commits, jingyue, arsenm
Differential Revision: http://reviews.llvm.org/D17316
llvm-svn: 261112
Justin Lebar [Wed, 17 Feb 2016 17:46:41 +0000 (17:46 +0000)]
Update langref to indicate that calls may be convergent.
Summary:
As previously written, only functions could be convergent. But calls
need to have a notion of convergence as well.
To see why this is important, consider an indirect call. We may or may
not want to disable optimizations around it and behave as though we're
calling a convergent function -- it depends on the semantics of the
language we're compiling. Thus the need for this attr on the call.
Reviewers: jingyue, joker.eph
Subscribers: llvm-commits, tra, jhen, arsenm, chandlerc, hfinkel, resistor
Differential Revision: http://reviews.llvm.org/D17314
llvm-svn: 261111
Justin Lebar [Wed, 17 Feb 2016 17:46:39 +0000 (17:46 +0000)]
Fix typo in comment.
llvm-svn: 261110
David Majnemer [Wed, 17 Feb 2016 17:19:00 +0000 (17:19 +0000)]
Correct more typos in conditional expressions
We didn't correctly handle some edge cases, causing us to bail out
before correcting all the typos.
llvm-svn: 261109
Chris Bieneman [Wed, 17 Feb 2016 16:57:38 +0000 (16:57 +0000)]
[CMake] [NFC] Move macro definitions out of config-ix.cmake
This change should have no functional impact, it just moves some macro definitions out of config-ix.cmake into CompilerRTUtils.cmake.
This step will allow these macros to be re-used by the separated builtin build.
llvm-svn: 261108
Rafael Espindola [Wed, 17 Feb 2016 16:48:00 +0000 (16:48 +0000)]
Represent the dynamic table itself with a DynRegionInfo.
The dynamic table is also an array of a fixed structure, so it can be
represented with a DynReginoInfo.
No major functionality change. The extra error checking is covered by
existing tests with a broken dynamic program header.
Idea extracted from r260488. I did the extra cleanups.
llvm-svn: 261107
Chris Bieneman [Wed, 17 Feb 2016 16:38:54 +0000 (16:38 +0000)]
[CMake] Push the dependency on AddLLVM into the test and unites layers
Compiler-rt only relies on LLVM for lit support. Pushing this dependency down into the test and unitest layers will allow builtin libraries to be built without LLVM.
llvm-svn: 261105
Mitch Bodart [Wed, 17 Feb 2016 16:35:18 +0000 (16:35 +0000)]
Fix some erroneous lit test failures due to unlucky name of working directory.
Differential Revision: http://reviews.llvm.org/D17044
llvm-svn: 261104
Rafael Espindola [Wed, 17 Feb 2016 16:21:49 +0000 (16:21 +0000)]
Add a unwrapOrError utility and use it to simplify ELFDumper.cpp.
Utility extracted from r260488.
llvm-svn: 261103
Samuel Benzaquen [Wed, 17 Feb 2016 16:13:14 +0000 (16:13 +0000)]
[clang-tidy] Match the type against the get() method we are calling,
instead of a get() method we find in the class.
The duck typed smart pointer class could have overloaded get() methods
and we should only skip the one that matches.
llvm-svn: 261102
Simon Pilgrim [Wed, 17 Feb 2016 15:52:39 +0000 (15:52 +0000)]
[X86][SSE] Update pshufb mask tests.
We are getting better at combining constant pshufb masks - use a real input instead of undef.
Add test for decoding multi-use bitcasted masks as well (actual support will come soon).
llvm-svn: 261101
Hongbin Zheng [Wed, 17 Feb 2016 15:49:21 +0000 (15:49 +0000)]
[Refactor] Move isl_ctx into Scop.
After we moved isl_ctx into Scop, we need to free the isl_ctx after
freeing all isl objects, which requires the ScopInfo pass to be freed
at last. But this is not guaranteed by the PassManager, and we need
extra code to free the isl_ctx at the right time.
We introduced a shared pointer to manage the isl_ctx, and distribute
it to all analyses that create isl objects. As such, whenever we free
an analyses with the shared_ptr (and also free the isl objects which
are created by the analyses), we decrease the (shared) reference
counter of the shared_ptr by 1. Whenever the reference counter reach
0 in the releaseMemory function of an analysis, that analysis will
be the last one that hold any isl objects, and we can safely free the
isl_ctx with that analysis.
Differential Revision: http://reviews.llvm.org/D17241
llvm-svn: 261100
Rafael Espindola [Wed, 17 Feb 2016 15:38:21 +0000 (15:38 +0000)]
Change how readobj stores info about dynamic symbols.
We used to keep both a section and a pointer to the first symbol.
The oddity of keeping a section for dynamic symbols is because there is
a DT_SYMTAB but no DT_SYMTABZ, so to print the table we have to find the
size via a section table.
The reason for still keeping a pointer to the first symbol is because we
want to be able to print relocation tables even if the section table is
missing (it is mandatory only for files used in linking).
With this patch we keep just a DynRegionInfo. This then requires
changing a few places that were asking for a Elf_Shdr but actually just
needed the first symbol.
The test change is to delete the program header pointer.
Now that we use the information of both DT_SYMTAB and .dynsym, we don't
depend on the sh_entsize of .dynsym if we see DT_SYMTAB.
Note: It is questionable if it is worth it putting the effort to report
broken sh_entsize given that in files with no section table we have to
assume it is sizeof(Elf_Sym), but that is for another change.
Extracted from r260488.
llvm-svn: 261099
Alexey Bataev [Wed, 17 Feb 2016 15:36:39 +0000 (15:36 +0000)]
[OPENMP] Fix tests incompatibility with ARM buildbots.
llvm-svn: 261098
Krzysztof Parzyszek [Wed, 17 Feb 2016 15:02:07 +0000 (15:02 +0000)]
[Hexagon] Fold object construction into map::insert
llvm-svn: 261096
Simon Pilgrim [Wed, 17 Feb 2016 14:56:58 +0000 (14:56 +0000)]
[X86][SSE] Update pshufb mask test to use a real input instead of undef
We are getting better at combining constant pshufb masks - this test would've failed once we decode bitcasted masks as well.
llvm-svn: 261095
Chad Rosier [Wed, 17 Feb 2016 14:45:36 +0000 (14:45 +0000)]
Typo.
llvm-svn: 261093
Igor Breger [Wed, 17 Feb 2016 14:04:33 +0000 (14:04 +0000)]
AVX512: Fix LowerMSCATTER() return value.
Bug description:
The bug was discovered when test was compiled with -O0.
In case scatter result is DAG root , VectorLegalizer failed (assert) due to LowerMSCATTER() return kmask as result.
Change LowerMSCATTER() to return chain as original node do.
Differential Revision: http://reviews.llvm.org/D17331
llvm-svn: 261090
Alexey Bataev [Wed, 17 Feb 2016 13:19:37 +0000 (13:19 +0000)]
[OPENMP 4.5] Codegen support for data members in 'firstprivate' clause.
Added codegen for captured data members in non-static member functions.
llvm-svn: 261089
Daniel Sanders [Wed, 17 Feb 2016 13:16:31 +0000 (13:16 +0000)]
[libcxx] Fix definition of regex_traits::__regex_word on big-endian glibc systems
Summary:
On glibc, the bits used for the various character classes is endian dependant
(see _ISbit() in ctypes.h) but __regex_word does not account for this and uses
a spare bit that isn't spare on big-endian. On big-endian, it overlaps with the
bit for graphic characters which causes '-', '@', etc. to be considered a word
character.
Fixed this by defining the value using _ISbit(15) on MIPS glibc systems. We've
restricted this to MIPS for now to avoid the risk of introducing failures in
other targets.
Fixes PR26476.
Reviewers: hans, mclow.lists
Subscribers: dsanders, cfe-commits
Differential Revision: http://reviews.llvm.org/D17132
llvm-svn: 261088
Simon Atanasyan [Wed, 17 Feb 2016 12:49:43 +0000 (12:49 +0000)]
[ELF][MIPS] Update test case expectations due changes in MIPS/MC
llvm-svn: 261085
Anastasia Stulova [Wed, 17 Feb 2016 11:34:37 +0000 (11:34 +0000)]
[OpenCL] Added half type literal with suffix h.
OpenCL Extension v1.2 s9.5 allows half precision floating point
type literals with suffices h or H when cl_khr_fp16 is enabled.
Example: half x = 1.0h;
Patch by Liu Yaxun (Sam)!
Differential Revision: http://reviews.llvm.org/D16865
llvm-svn: 261084
Scott Egerton [Wed, 17 Feb 2016 11:15:16 +0000 (11:15 +0000)]
[mips] Removed the SHF_ALLOC flag and the SHT_REL flag from the .pdr section.
This section is used for debug information and has no need to be
in memory at runtime. This patch also fixes an error when compiling
the Linux kernel. The error is that there are relocations within the
.pdr section in a VDSO. SHT_REL was removed as it is a section type
and not a section flag, therefore it does not make sense for it to
be there. With this patch, LLVM now emits the same flags as
the GNU assembler.
llvm-svn: 261083
Simon Pilgrim [Wed, 17 Feb 2016 10:50:06 +0000 (10:50 +0000)]
[X86][AVX] Support bit-blend integer shuffles for 256-bit integer vectors
AVX1 doesn't support the shuffling of 256-bit integer vectors. For 32/64-bit elements we get around this by shuffling as float/double but for 8/16-bit elements (assuming they can't widen) we currently just split, shuffle as 128-bit vectors and concatenate the results back.
This patch adds the ability to lower using the bit-blend patterns before defaulting to the splitting behaviour.
Part 2 of 2
Differential Revision: http://reviews.llvm.org/D17292
llvm-svn: 261082
Simon Pilgrim [Wed, 17 Feb 2016 10:37:49 +0000 (10:37 +0000)]
[X86][AVX] Support bit-mask integer shuffles for 256-bit integer vectors
AVX1 doesn't support the shuffling of 256-bit integer vectors. For 32/64-bit elements we get around this by shuffling as float/double but for 8/16-bit elements (assuming they can't widen) we currently just split, shuffle as 128-bit vectors and concatenate the results back.
This patch adds the ability to lower using the bit-mask patterns before defaulting to the splitting behaviour. In some cases this ends up matching what AVX2 would do anyhow or what AVX1 does on the split vectors.
Part 1 of 2
Differential Revision: http://reviews.llvm.org/D17292
llvm-svn: 261081
Alexey Bataev [Wed, 17 Feb 2016 10:29:05 +0000 (10:29 +0000)]
[OPENMP] Fix handling loop-based directives with arrays.
Patch fixes possible problems with correct handling arrays as
expressions in initialization, conditions etc in loop-based constructs.
llvm-svn: 261080
Simon Pilgrim [Wed, 17 Feb 2016 10:12:30 +0000 (10:12 +0000)]
[X86][SSE] Tidyup BUILD_VECTOR operand collection. NFCI.
Avoid reuse of operand variables, keep them local to a particular lowering - the operand collection is unique to each case anyhow.
Renamed from V to Ops to more closely match their purpose.
llvm-svn: 261078
Benjamin Kramer [Wed, 17 Feb 2016 09:28:45 +0000 (09:28 +0000)]
[Hexagon] cast<> a reference instead of referencing + dereferencing.
llvm-svn: 261077
Jonas Hahnfeld [Wed, 17 Feb 2016 07:12:18 +0000 (07:12 +0000)]
[compiler-rt][msan] Ensure initialisation before calling __msan_unpoison
__msan_unpoison uses intercepted memset which currently leads to a SEGV
when linking with libc++ under CentOS 7.
Differential Revision: http://reviews.llvm.org/D17263
llvm-svn: 261073
David Blaikie [Wed, 17 Feb 2016 07:00:24 +0000 (07:00 +0000)]
llvm-dwp: Support for type units when merging DWPs into larger DWPs
llvm-svn: 261072
David Blaikie [Wed, 17 Feb 2016 07:00:22 +0000 (07:00 +0000)]
Fix the hash function.
llvm-svn: 261071
Cong Hou [Wed, 17 Feb 2016 06:37:04 +0000 (06:37 +0000)]
Detecte vector reduction operations just before instruction selection.
This patch detects vector reductions before instruction selection. Vector
reductions are vectorized reduction operations, and for such operations we have
freedom to reorganize the elements of the result as long as the reduction of them
stay unchanged. This will enable some reduction pattern recognition during
instruction combine such as SAD/dot-product on X86. A flag is added to
SDNodeFlags to mark those vector reduction nodes to be checked during instruction
combine.
To detect those vector reductions, we search def-use chains starting from the
given instruction, and check if all uses fall into two categories:
1. Reduction with another vector.
2. Reduction on all elements.
in which 2 is detected by recognizing the pattern that the loop vectorizer
generates to reduce all elements in the vector outside of the loop, which
includes several ShuffleVector and one ExtractElement instructions.
Differential revision: http://reviews.llvm.org/D15250
llvm-svn: 261070
Rui Ueyama [Wed, 17 Feb 2016 06:08:42 +0000 (06:08 +0000)]
Make getOffset a member function of DynamicReloc<ELFT>.
Logically it belongs to DynamicReloc, and it is more readable to
be a member of the class.
llvm-svn: 261069
Rui Ueyama [Wed, 17 Feb 2016 05:40:03 +0000 (05:40 +0000)]
Use shorter names for the .gnu.hash class.
llvm-svn: 261067
Rui Ueyama [Wed, 17 Feb 2016 05:40:01 +0000 (05:40 +0000)]
Use stable_partition instead of erasing all elements and fill it again.
llvm-svn: 261066
Rui Ueyama [Wed, 17 Feb 2016 05:06:40 +0000 (05:06 +0000)]
Use an accurate type instead of unsigned.
These values are offsets in the string table (which must fit in
host computer's memory space), so size_t is better than unsigned.
llvm-svn: 261065
Rui Ueyama [Wed, 17 Feb 2016 04:56:44 +0000 (04:56 +0000)]
Split SymbolTableSection::writeGlobalSymbols.
Previously, we added garbage-collected symbols to the symbol table
and filter them out when we were writing symbols to the file. In
this patch, garbage-collected symbols are filtered out from beginning.
llvm-svn: 261064
Hans Wennborg [Wed, 17 Feb 2016 02:49:59 +0000 (02:49 +0000)]
Revert r260979 "[X86] Enable the LEA optimization pass by default."
Asserts are still firing in Chromium builds. PR26575.
llvm-svn: 261058
Xinliang David Li [Wed, 17 Feb 2016 02:39:34 +0000 (02:39 +0000)]
revert r261038: arm/aarch64 bot failure
llvm-svn: 261057
Mehdi Amini [Wed, 17 Feb 2016 02:18:58 +0000 (02:18 +0000)]
Revert "Query the StringMap only once when creating MDString (NFC)"
This reverts commit r261030 and r261036.
(The revision was marked "approved" on phabricator, but some concerns
were raised on the mailing list. Thanks D. Blaikie for notifying me.)
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261055
Chandler Carruth [Wed, 17 Feb 2016 02:13:35 +0000 (02:13 +0000)]
[cmake] Revert r260742 (and r260744) to improve order file support.
This appears to be passing '-Wl,-order_file' to Linux link commands,
which then causes the linker to silently, behind the scenes, write the
output to 'rder_file' instead of somewhere else. Will work with Chris to
figure out the proper support for this, but so far there are numerous
people who can't get Clang to update when they build because of this.
llvm-svn: 261054
Sean Silva [Wed, 17 Feb 2016 02:08:19 +0000 (02:08 +0000)]
[AttrDocs.td] Fix up some reST syntax.
llvm-svn: 261053
Haicheng Wu [Wed, 17 Feb 2016 02:01:50 +0000 (02:01 +0000)]
[AliasSetTracker] Teach AliasSetTracker about MemSetInst
This change is to fix the problem discussed in
http://lists.llvm.org/pipermail/llvm-dev/2016-February/095446.html.
llvm-svn: 261052
JF Bastien [Wed, 17 Feb 2016 01:59:23 +0000 (01:59 +0000)]
WebAssembly: update expected failures
r261050 seems to inadvertently fix the assertion failure.
llvm-svn: 261051
Dan Gohman [Wed, 17 Feb 2016 01:43:37 +0000 (01:43 +0000)]
[WebAssembly] Call memcpy for large byval copies.
This fixes very slow compilation on
test/CodeGen/Generic/2010-11-04-BigByval.ll . Note that MaxStoresPerMemcpy
and friends are not yet carefully tuned so the cutoff point is currently
somewhat arbitrary. However, it's important that there be a cutoff point
so that we don't emit unbounded quantities of loads and stores.
llvm-svn: 261050
Evgeniy Stepanov [Wed, 17 Feb 2016 01:34:56 +0000 (01:34 +0000)]
[msan] Extend prlimit test.
llvm-svn: 261049
Evgeniy Stepanov [Wed, 17 Feb 2016 01:26:57 +0000 (01:26 +0000)]
[msan] Intercept prlimit.
llvm-svn: 261048
Xinliang David Li [Wed, 17 Feb 2016 00:59:01 +0000 (00:59 +0000)]
Test simplification
llvm-svn: 261047
Xinliang David Li [Wed, 17 Feb 2016 00:58:13 +0000 (00:58 +0000)]
Restrengthen tests relaxed in r259955
llvm-svn: 261046
Mehdi Amini [Wed, 17 Feb 2016 00:42:20 +0000 (00:42 +0000)]
Teach clang to use the ThinLTO pipeline
Summary: Use the new pipeline implemented in D17115
Reviewers: tejohnson
Subscribers: joker.eph, cfe-commits
Differential Revision: http://reviews.llvm.org/D17272
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261045
JF Bastien [Wed, 17 Feb 2016 00:34:15 +0000 (00:34 +0000)]
WebAssembly: update expected test failures
r261032 adds frame address support.
llvm-svn: 261044
Matt Arsenault [Wed, 17 Feb 2016 00:27:31 +0000 (00:27 +0000)]
Add .gitignore for build directories
llvm-svn: 261043
Matt Arsenault [Wed, 17 Feb 2016 00:27:27 +0000 (00:27 +0000)]
amdgcn: Use new workitem intrinsics
llvm-svn: 261042
Chandler Carruth [Wed, 17 Feb 2016 00:18:16 +0000 (00:18 +0000)]
[LCG] Construct an actual call graph with call-edge SCCs nested inside
reference-edge SCCs.
This essentially builds a more normal call graph as a subgraph of the
"reference graph" that was the old model. This allows both to exist and
the different use cases to use the aspect which addresses their needs.
Specifically, the pass manager and other *ordering* constrained logic
can use the reference graph to achieve conservative order of visit,
while analyses reasoning about attributes and other properties derived
from reachability can reason about the direct call graph.
Note that this isn't necessarily complete: it doesn't model edges to
declarations or indirect calls. Those can be found by scanning the
instructions of the function if desirable, and in fact every user
currently does this in order to handle things like calls to instrinsics.
If useful, we could consider caching this information in the call graph
to save the instruction scans, but currently that doesn't seem to be
important.
An important realization for why the representation chosen here works is
that the call graph is a formal subset of the reference graph and thus
both can live within the same data structure. All SCCs of the call graph
are necessarily contained within an SCC of the reference graph, etc.
The design is to build 'RefSCC's to model SCCs of the reference graph,
and then within them more literal SCCs for the call graph.
The formation of actual call edge SCCs is not done lazily, unlike
reference edge 'RefSCC's. Instead, once a reference SCC is formed, it
directly builds the call SCCs within it and stores them in a post-order
sequence. This is used to provide a consistent platform for mutation and
update of the graph. The post-order also allows for very efficient
updates in common cases by bounding the number of nodes (and thus edges)
considered.
There is considerable common code that I'm still looking for the best
way to factor out between the various DFS implementations here. So far,
my attempts have made the code harder to read and understand despite
reducing the duplication, which seems a poor tradeoff. I've not given up
on figuring out the right way to do this, but I wanted to wait until
I at least had the system working and tested to continue attempting to
factor it differently.
This also requires introducing several new algorithms in order to handle
all of the incremental update scenarios for the more complex structure
involving two edge colorings. I've tried to comment the algorithms
sufficiently to make it clear how this is expected to work, but they may
still need more extensive documentation.
I know that there are some changes which are not strictly necessarily
coupled here. The process of developing this started out with a very
focused set of changes for the new structure of the graph and
algorithms, but subsequent changes to bring the APIs and code into
consistent and understandable patterns also ended up touching on other
aspects. There was no good way to separate these out without causing
*massive* merge conflicts. Ultimately, to a large degree this is
a rewrite of most of the core algorithms in the LCG class and so I don't
think it really matters much.
Many thanks to the careful review by Sanjoy Das!
Differential Revision: http://reviews.llvm.org/D16802
llvm-svn: 261040
Reid Kleckner [Wed, 17 Feb 2016 00:17:33 +0000 (00:17 +0000)]
[X86] Fix a shrink-wrapping miscompile around __chkstk
__chkstk clobbers EAX. If EAX is live across the prologue, then we have
to take extra steps to save it. We already had code to do this if EAX
was a register parameter. This change adapts it to work when shrink
wrapping is used.
llvm-svn: 261039
Xinliang David Li [Wed, 17 Feb 2016 00:14:52 +0000 (00:14 +0000)]
New test case: make sure alloc bit is not set for covmap section on Linux
llvm-svn: 261038
Dan Gohman [Wed, 17 Feb 2016 00:14:03 +0000 (00:14 +0000)]
[WebAssembly] Use SDValue::getConstantOperandVal. NFC.
llvm-svn: 261037
Mehdi Amini [Wed, 17 Feb 2016 00:11:59 +0000 (00:11 +0000)]
Fix MSVC bot: apparently visual studio does not like explicitly defaulted move ctor
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261036
Richard Smith [Wed, 17 Feb 2016 00:04:04 +0000 (00:04 +0000)]
Improve diagnostics for ill-formed literal operator declarations.
Patch by Erik Pilkington!
llvm-svn: 261034
Andrew Kaylor [Tue, 16 Feb 2016 23:52:18 +0000 (23:52 +0000)]
Fix build LLVM with -D LLVM_USE_INTEL_JITEVENTS:BOOL=ON on Windows
Differential Revision: http://reviews.llvm.org/D16940
llvm-svn: 261033
Dan Gohman [Tue, 16 Feb 2016 23:48:04 +0000 (23:48 +0000)]
[WebAssembly] Implement __builtin_frame_address.
Differential Revision: http://reviews.llvm.org/D17307
llvm-svn: 261032
Mehdi Amini [Tue, 16 Feb 2016 23:05:56 +0000 (23:05 +0000)]
Query the StringMap only once when creating MDString (NFC)
Summary: Loading IR with debug info improves MDString::get() from 19ms to 10ms.
Reviewers: dexonsmith
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D16597
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261030
Mehdi Amini [Tue, 16 Feb 2016 23:02:29 +0000 (23:02 +0000)]
Define the ThinLTO Pipeline (experimental)
Summary:
On the contrary to Full LTO, ThinLTO can afford to shift compile time
from the frontend to the linker: both phases are parallel (even if
it is not totally "free": projects like clang are reusing product
from the "compile phase" for multiple link, think about
libLLVMSupport reused for opt, llc, etc.).
This pipeline is based on the proposal in D13443 for full LTO. We
didn't move forward on this proposal because the LTO link was far too
long after that. We believe that we can afford it with ThinLTO.
The ThinLTO pipeline integrates in the regular O2/O3 flow:
- The compile phase perform the inliner with a somehow lighter
function simplification. (TODO: tune the inliner thresholds here)
This is intendend to simplify the IR and get rid of obvious things
like linkonce_odr that will be inlined.
- The link phase will run the pipeline from the start, extended with
some specific passes that leverage the augmented knowledge we have
during LTO. Especially after the inliner is done, a sequence of
globalDCE/globalOpt is performed, followed by another run of the
"function simplification" passes. It is not clear if this part
of the pipeline will stay as is, as the split model of ThinLTO
does not allow the same benefit as FullLTO without added tricks.
The measurements on the public test suite as well as on our internal
suite show an overall net improvement. The binary size for the clang
executable is reduced by 5%. We're still tuning it with the bringup
of ThinLTO and it will evolve, but this should provide a good starting
point.
Reviewers: tejohnson
Differential Revision: http://reviews.llvm.org/D17115
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261029
Mehdi Amini [Tue, 16 Feb 2016 22:54:27 +0000 (22:54 +0000)]
Refactor the PassManagerBuilder: extract a "addFunctionSimplificationPasses()" (NFC)
It is intended to contains the passes run over a function after the
inliner is done with a function and before it moves to its callers.
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261028
Adam Nemet [Tue, 16 Feb 2016 22:50:19 +0000 (22:50 +0000)]
Fix test from r261013
llvm-svn: 261027
Simon Pilgrim [Tue, 16 Feb 2016 22:33:27 +0000 (22:33 +0000)]
[X86][AVX] Regenerated vselect tests
llvm-svn: 261026
Ahmed Bougacha [Tue, 16 Feb 2016 22:14:12 +0000 (22:14 +0000)]
[X86] Remove the now-unused X86ISD::PSIGN. NFC.
llvm-svn: 261025
Ahmed Bougacha [Tue, 16 Feb 2016 22:14:07 +0000 (22:14 +0000)]
[X86] Generalize logic blend of (x, -x) combine to match (-x, x).
I suspect this is what let PR26110 lie dormant for so long.
llvm-svn: 261024
Ahmed Bougacha [Tue, 16 Feb 2016 22:14:03 +0000 (22:14 +0000)]
[X86] Don't turn (c?-v:v) into (c?-v:0) by blindly using PSIGN.
Currently, we sometimes miscompile this vector pattern:
(c ? -v : v)
We lower it to (because "c" is <4 x i1>, lowered as a vector mask):
(~c & v) | (c & -v)
When we have SSSE3, we incorrectly lower that to PSIGN, which does:
(c < 0 ? -v : c > 0 ? v : 0)
in other words, when c is either all-ones or all-zero:
(c ? -v : 0)
While this is an old bug, it rarely triggers because the PSIGN combine
is too sensitive to operand order. This will be improved separately.
Note that the PSIGN tests are also incorrect. Consider:
%b.lobit = ashr <4 x i32> %b, <i32 31, i32 31, i32 31, i32 31>
%sub = sub nsw <4 x i32> zeroinitializer, %a
%0 = xor <4 x i32> %b.lobit, <i32 -1, i32 -1, i32 -1, i32 -1>
%1 = and <4 x i32> %a, %0
%2 = and <4 x i32> %b.lobit, %sub
%cond = or <4 x i32> %1, %2
ret <4 x i32> %cond
if %b is zero:
%b.lobit = <4 x i32> zeroinitializer
%sub = sub nsw <4 x i32> zeroinitializer, %a
%0 = <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
%1 = <4 x i32> %a
%2 = <4 x i32> zeroinitializer
%cond = or <4 x i32> %a, zeroinitializer
ret <4 x i32> %a
whereas we currently generate:
psignd %xmm1, %xmm0
retq
which returns 0, as %xmm1 is 0.
Instead, use a pure logic sequence, as described in:
https://graphics.stanford.edu/~seander/bithacks.html#ConditionalNegate
Fixes PR26110.
Differential Revision: http://reviews.llvm.org/D17181
llvm-svn: 261023
Ahmed Bougacha [Tue, 16 Feb 2016 22:13:59 +0000 (22:13 +0000)]
[X86] Extract PSIGN/BLENDVP tests into vector-blend.ll. NFC.
We're going to stop generating PSIGN, so calling a test "psign"
isn't ideal. Instead, call these tests what they really are:
variable blends using logic.
Also add a test to exhibit a case we're currently missing in
the PSIGN combine.
llvm-svn: 261022
Ahmed Bougacha [Tue, 16 Feb 2016 22:13:55 +0000 (22:13 +0000)]
[X86] Extract PSIGN/BLENDVP combine. NFC.
llvm-svn: 261021
Ahmed Bougacha [Tue, 16 Feb 2016 22:13:49 +0000 (22:13 +0000)]
[X86] Extract ANDNP combine. NFC.
This makes it IMO more readable and reduces indentation.
llvm-svn: 261020
Mehdi Amini [Tue, 16 Feb 2016 22:07:03 +0000 (22:07 +0000)]
Bitcode writer: fix a typo, using getName() instead of getSourceFileName()
When emitting the source filename, the encoding of the string
was checked against the name instead of the filename.
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 261019
Artem Belevich [Tue, 16 Feb 2016 22:03:20 +0000 (22:03 +0000)]
[CUDA] pass debug options to ptxas.
ptxas optimizations are disabled if we need to generate debug info
as ptxas does not accept '-g' otherwise.
Differential Revision: http://reviews.llvm.org/D17111
llvm-svn: 261018
Derek Schuff [Tue, 16 Feb 2016 21:52:06 +0000 (21:52 +0000)]
[WebAssembly] Update torture test expectations
These were fixed with r260978
llvm-svn: 261017